Journal of Sound and Vibration (2003) 259(4), 917933

doi:10.1006/jsvi.2002.5129, available online at http://www.idealibrary.com on

APPROXIMATE ANALYTICAL SOLUTIONS FOR PRIMARY

CHATTER IN THE NON-LINEAR METAL CUTTING
MODEL
J. Warmi N! ski and G. Litak
Department of Applied Mechanics, Technical University of Lublin, Nadbystrzycka 36,
Lublin 20-618, Poland

M. P. Cartmell
Department of Mechanical Engineering, James Watt Building, University of Glasgow, Glasgow,
G12 8QQ, Scotland, UK. E-mail: matthewc@mech.gla.ac.uk

R. Khanin
Department of Applied Mathematics and Theoretical Physics, University of Cambridge,
Silver St., Cambridge, CB3 9EW, England, UK

and
M. Wiercigroch
Centre for Applied Dynamics Research, Department of Engineering, Fraser Noble Building,
Kings College, University of Aberdeen, Aberdeen, AB24 3UE, Scotland, UK
(Received 2 October 2001, and in final form 8 April 2002)
This paper considers an accepted model of the metal cutting process dynamics in the
context of an approximate analysis of the resulting non-linear differential equations of
motion. The process model is based upon the established mechanics of orthogonal cutting
and results in a pair of non-linear ordinary differential equations which are then restated in
a form suitable for approximate analytical solution. The chosen solution technique is the
perturbation method of multiple time scales and approximate closed-form solutions are
generated for the most important non-resonant case. Numerical data are then substituted
into the analytical solutions and key results are obtained and presented. Some comparisons
between the exact numerical calculations for the forces involved and their reduced and
simplied analytical counterparts are given. It is shown that there is almost no discernible
difference between the two thus conrming the validity of the excitation functions adopted
in the analysis for the data sets used, these being chosen to represent a real orthogonal
cutting process. In an attempt to provide guidance for the selection of technological
parameters for the avoidance of primary chatter, this paper determines for the rst time the
stability regions in terms of the depth of cut and the cutting speed co-ordinates.
# 2002 Elsevier Science Ltd. All rights reserved.

1. INTRODUCTION

The mechanics of metal cutting has been of interest to a large number of researchers for
some considerable time and early models for the intermittent cutting process were
proposed by Tobias [1], Opitz and Bernardi [2], and Tlusty [3]. These models were useful in
that they attempted to incorporate the necessary physical interactions between the
0022-460X/02/$35.00

# 2002 Elsevier Science Ltd. All rights reserved.

918

J. WARMINSKI ET AL.

mechanics of the machine tool and the actual cutting process itself. Grabec [4, 5] provided
considerable insight into this interactive approach to the modelling of self-excited chatter
vibrations by providing useful numerical solutions and interpretations of practical
problem data. Marui et al. [6], Lin and Weng [7, 8], Gradisek et al. [911], Stepan and
Kalmar-Nagy [12], Litak et al. [13], Przystupa and Litak [14], Johnson and Moon [15],
Stepan [16], Wiercigroch and Budak [17], and Moon and Kalmar-Nagy [18] carried out
further work on this general theme. Rabinowicz [19], Bell and Burdekin [20] and
Wiercigroch [21, 22] considered an additional phenomenological issue, in the form of
stickslip effects observable in the machine tool slide-way.
Chaotic motions have been observed in the responses of chattering machine-tool
systems by Grabec [4, 5] and Wiercigroch [22, 23], Wiercigroch and Cheng [24], Warmin! ski
et al. [25], Kalmar-Nagy and Pratt [26], and Wiercigroch and Krivtsov [27]. These authors
have examined the dynamics of several different variations on the basic tool-to-work-piece
conguration. The general conclusion to be drawn from their work is that intermittency
(i.e., chatter vibration) is consistently seen to be responsible for unstable and bifurcatory
behaviour for certain sets of design data. In other words, it seems possible to nd practical
design data for a large variety of models and systems which will lead, ultimately, to
instability, and to bifurcations leading to chaos. This in itself is an important nding
because it indicates that the metal cutting process is potentially relatively difcult to
control, and that stable cutting cannot necessarily be guaranteed for all system designs. It
also implies the possibility of critical performance degradation in certain cases of metal
cutting. Practically speaking, stickslip and intermittency effects may arise because of
deviations in the work-piece, the jamming up of chip formation, or dynamical delay effects
due to chip thickness variation causing unevenness in the work-piece surface.
Approximate analytical solutions have been proposed for modelling different aspects of
the non-linear dynamical behaviour of the cutting process in recent years. For example, a
non-linear one-degree-of-freedom cutting system has been investigated by Nosyreva and
Molinari [28] and analytical solutions were obtained by applying the multiple scales method.
In this work the main inuence was of the cutting velocity on the mean friction coefcient,
thereby reecting the mechanics of the Merchant model [29]. The problem was reduced to a
consideration of motion in the direction perpendicular to the cutting velocity. In addition to
this the inuence of the ploughing force on the system dynamics was analyzed, showing that
for some supercritical conditions the amplitude of vibration saturates due to non-linear
effects. This has been conrmed in a slightly different model by Wiercigroch and Krivtsov
[27], for the case where the discontinuity of the friction characteristics is responsible for
bounding the oscillations. A two-dimensional dynamical investigation of orthogonal cutting
was carried out by Hwang et al. [30], in which a simple model of chip formation and the
regenerative cutting effects was analyzed using the multiple scales method. This approach
considered the system under investigation to be weakly non-linear and therefore assessed the
inuence of strong non-linearities by direct numerical integration. This numerical study
revealed that a different cutting thicknesses can result in qualitatively different behaviour of
the system. A non-linear single-degree-of-freedom model originally proposed by Hanna and
Tobias [31] has been discussed by Nayfeh et al. [32]. This model included linear structural
damping, quadratic and cubic non-linear stiffness of the machine tool, and linear, quadratic
and cubic regenerative terms, as required to dene the regenerative cutting process in
dynamical terms. The method of multiple scales and bifurcation analysis were both applied
to the problem in order to predict the complex dynamics of the system. By using phase
portraits, Poincar!e sections, and power spectra, periodic, quasiperiodic, and chaotic
attractors were constructed and the importance of including non-linear stiffness terms was
also emphasized.

NON-LINEAR METAL CUTTING MODEL

919

As shown in the discussion above there is evidence of growing interest in the application
of approximate analytical methods to support numerical analysis of the dynamic
interactions occurring in cutting processes, and in particular, to provide guidance on how
to select technological parameters in order avoid primary chatter vibrations. This paper
looks, for the rst time, at the stability regions in the context of the depth of cut and the
cutting speed co-ordinates, and in order to achieve this the principal foundation of the
work has been based on an investigation of the original dynamical model proposed by
Wiercigroch [23], noting that this model has been derived from that of Grabec [4, 5]. In
this paper, Wiercigrochs model has been developed to include the effects of Coulomb
friction, but in the context of analytical rather than numerical solutions. Naturally, this
pre-supposes the existence of periodic solution forms both for the general non-resonant
condition as well as certain internal resonance conditions, but obviously will not deal with
cases of non-periodic responses. The non-resonant condition is interesting because, in
terms of the frequency domain, it lies between and around the various resonances that the
multiple scales analysis also predicts. Therefore, a non-resonant response is a non-trivial
vibration which is spread across the working frequency domain, but with the various
resonances taking precedence at appropriate points in the domain. In this paper the nonresonant chatter vibration is analyzed to show conditions in which it can grow and vice
versa. The main resonant cases are stated but not analyzed here. The perturbation
technique of multiple scales is applied to the problem, after some manipulation of the basic
system equations of motion. Bespoke computer algebra software is used to generate the
perturbation equations (see references [33, 34]) and to provide the required analytical
solutions to a pre-specied level of approximation accuracy. It is also shown that the
analytically complicated equations of motion due to Wiercigroch [23] can be simplied
somewhat without too much loss of generality or accuracy. This simplication applies
specically to the orthogonal cutting forces fx and fy and is discussed further within the
following analysis.

2. MODELLING OF SYSTEM DYNAMICS

On the basis of the work of Juilian et al. [35], in which it was shown that there is inherent
discontinuity in the metal cutting process, and the observation that there is feedback
between the machine tool and the actual cutting process (in the form of a vector of cutting
forces) Wiercigroch [23] proposed a revised version of the model originally suggested by
Grabec [4, 5]. This revision to the original work of Grabec [4, 5] is discussed in detail in
reference [23] in the form of additional frictional modelling. This is related to the work of
Hastings et al. [36] in which it was shown that there is a physically based mathematical
interdependence between the orthogonal cutting force terms, noting that these terms are
made up of discontinuous frictional effects. The ensuing model is considered to be
reasonably representative of the process for this reason and also because of its adoption of
orthogonal machining whereby the cutting edge is parallel to the work-piece and normal to
the cutting direction as shown in Figure 1.
The result of this process of model development is a two-degree-of-freedom
representation in which Heaviside functions appear within the frictionally based force
terms fx and fy : These are necessary to represent the stickslip discontinuity, as is the
presence of the relative velocity vr between the tool and the work-piece in the x direction
(the direction of travel of the work-piece). The two degrees of freedom are represented by
x and y; which are non-dimensionalized generalized co-ordinates dening motion of the
tool in the x and the y directions (the latter being axially along the tool). The tool is

920

J. WARMINSKI ET AL.

Figure 1. Schematic diagram of the orthogonal metal cutting process, according to Wiercigroch [23].

considered to be supported by means of foundation-mounted linear springs and dampers

oriented in both directions, a relatively straightforward but not unrealistic assumption. In
this way, the elastic and dissipative properties are dened for the tool and its mounting
assembly. This is therefore a good denitive, and practically tractable, model for the
process and in the form suggested by Wiercigroch [23] it can be seen that the equations of
motion for the system are given as follows:



1
m0
2
00
0
x 2xx x x q0 hC1 jvr j  1 1 Hh Hvr
sgnvr
fx ; 1
1 m0
1 m0
 
p
y00 2xy ay0 ay C2 vf   12 1 C3 h  12 1 Hfx sgnvf q0 hC1 jvr j  12 1



1
m0
sgnvr
2
Hh Hvr
fy ;
1 mo
1 m0
The right-hand side terms in each equation collectively form the non-dimensionalized
forces fx and fy : It should also be noted that q0 is the magnitude of the cutting force, h0 is
initial depth of cut, m0 is the static friction coefcient, and that the relative velocity
between the tool and the work-piece in the x and y directions, respectively, vr and vf ; and
the actual depth of cut, h; are dened and related as follows, given that v0 ; R; R0 and C14
are the absolute velocity of the work-piece, the coefcient of variable shear plastic
deformation, the shear plastic deformation constant, and cutting process constants, thus,
vr v0  x 0 ;
h h0  y:

vf v0  Ry0 v0  R0 C4 vr  12 1 y0 v0  R0 C4 v0  x0  1 1 y0 ;

NON-LINEAR METAL CUTTING MODEL

921

Equation (1) can be rewritten in simpler form by rstly substituting the forms given above
for the relative velocities in x and y and the depth of cut (vr ; vf ; and h) and then lumping
together the terms that emerge from this so that the coefcients in the differential equation
are simplied as far as possible, in the limit of small oscillations,
x00 2xx b2 x0 x b3 y  b1 x02 b4 x02 y  b5 x0 y Q2 ;

where



1
b3 Q1 C1 v0  12
;
C1


1
m
sgnvr 0 Hh;
b4 Q1 C1 ; b5 2Q1 C1 v0  1; Q1 q0 Hvr
1 mo
1 m0


2
Q2 Q1 h0 C1 v0  2h0 C1 v0 h0 C1 h0 :
b1 Q1 h0 C1 ;

b2 2Q1 ho C1 v0  1;

In the case of equation (2) certain groupings of constants can be replaced by simple
substitutions to lead to a differential equation which is more explicit (note that this is also
because of the use of the substitutions given above for vr ; vf ; and h once again), therefore,
p
y00 2xy ay0 ay Q3 h0  yC1 s2 C1 x02  2C1 v0 x0 2C1 x0 1 C2 d21 2C2 d1 d2 x02 y0
2C2 d1 d3 x0 y0  2C2 d1 d4 y0 C2 d22 x04 y02  2C2 d2 d3 x03 y02
2C2 d2 d4 x02 y02 C2 d23 x02 y02  2C2 d3 d4 x0 y02 C2 d24 y02 1


C3 s1 C3 y2 2C3 y  2C3 h0 y 1 ;

where
Q3 Hfx sgnvf Q1 ;
d1 v0  1;

s1 h20  2h0 1;

s2 v20  2v0 1;

d2 R0 C4 ; d3 2R0 C4 v0  1;

d4 R0 C4 v20  2C4 v0 C4 1:

Equation (4) cannot be used directly as it stands, and needs further expansion of the righthand side. However, this is problematic because of the large number of terms that are
generated when this is done. To combat this a perturbation ordering scheme is proposed at
this stage. The ordering procedure is based upon introducing an (assumed small)
perturbation parameter e into the machining constants C1 ; C2 ; C3 ; and C4 ; and also into
the two damping parameters for the tool, xx and xy : It is relevant to note that this
automatically assists in the denitions of certain grouped constants into single constant
terms P1 and P2 : The outcome of this is that terms of order greater than Oe1 disappear
from equation (4) and all remaining non-linear terms in that equation go to Oe1 : No
higher order terms are generated by also applying this procedure to equation (3), so all
terms from that equation are naturally retained. The physical basis for this ordering
protocol is that the effect of the numerical values used for C1 ; C2 ; C3 ; and C4 (see reference
[23]), when coupled with other terms, is to make them appear as relatively small terms to
Oe2 ; or higher order, in the differential equations. It is important to appreciate that the
parameters C1 ; C2 ; C3 ; and C4 inuence the functions fx and fy and that these parameters
are coupled with other parameters such as cutting velocity and cutting depth. The data
used for C1 ; C2 ; C3 ; and C4 has been taken from the previously published work of Grabec
[5], and Wiercigroch [23], both of which were reliant on experimental data. The form of the
resulting expansion of equation (4) is also far more algebraically manageable than it
otherwise would have been. Damping can be assumed to be to Oe1 ; in line with
experimental ndings. The additional grouped constants (denoted by the overbars) are
dened in an attempt to tidy up the right-hand side terms in both differential equations,

922

J. WARMINSKI ET AL.

whilst also retaining explicit ordering for the perturbation analysis that comes next,
C1 eC1 ;

C2 eC2 ;

C3 eC3 ;

xx enx ;

xy eny ;

P1 eP1 2C3 Q3 h0 1  h0 C1 Q3 s2 C3 Q3 s1 C2 Q3 v20  2C2 Q3 v0 C2 Q3 ;

P2 eP2 C1 Q3 h0 s2 C3 Q3 h0 s1 C2 Q3 h0 v20  2C2 Q3 h0 v0 C2 Q3 h0 ;
g1 2Q1 h0 C1 v0  1;
g 4 Q1 C1 ;

g3 Q1 h0 C1 ;

g5 2Q1 C1 v0  1; l1 2C2 Q3 h0 R0 1  v0 ;

l2 C2 Q3 h0 R20 ;
l 5 C3 Q3 ;

g2 Q1 C1 v0  12 ;

l3 C 1 Q 3 ;

l4 2C1 Q3 1  v0 ;

l6 2C2 Q3 R0 1  v0 ;

l7 C2 Q3 R20 ;

l9 2C1 Q3 h0 1  v0 ; l10 C3 Q3 3h0  2;

l8 C1 Q3 h0 ;

l11 Q3 h0 :

Equations (5) and (6), given below, are the nal forms of the system equations and imply a
perturbation scheme of up to and including rst order e. Substitution of the above forms lead
to the emergence of these two structurally simplied equations of motion for the system,
x00 x e2nx g1 x0  eg2 y eg3 x02  eg4 x02 y eg5 x0 y e0 Q2  e0 Q1 y;

p
y00 a Q3 y e2ny a l1 y0  eP% 1 y el2 y02  el3 x02 y  el4 x0 y  el5 y3  el6 yy0
el7 yy02 el8 x02 el9 x0 el10 y2 eP2 e0 l11 :

The consequence of neglecting terms from the right-hand side of equation (4) to get equation
(6) has now to be discussed despite the implication of relative smallness of these terms by the
chosen ordering protocol. The right-hand side function fx in equation (1) should be
compared with the right-hand side of equation (5) (but with the term 2nx removed), and,
correspondingly, function fy in equation (2) should be compared withp
the
right-hand side of
equation (6) (noting that the term 2Q3 y must be added and term 2n a must be removed).
On that basis the functions describing the cutting process can be compared for the full and
the simplied equations. This has been investigated for many parameter values with little
difference evident for those cases. It was decided from thereon to restrict the comparison in
the paper to the case based on the data of Grabec [4, 5] and Wiercigroch [23]. In Figures 2(a)
and 2(b) the solid curves represent the analytical results obtained from the multiple scales
analysis, whereas the dotted curves are obtained from numerical integration of the original
equations (1) and (2). This shows that there are no visible differences between the two forms
for fx and only the most minor differences evident between the two forms for fy :
This is for typical cutting data and also for the two forces as functions of vr and vf ;
respectively (Figures 2(a) and 2(b) and then as functions of h (Figures 2(c) and 2(d). This
provides sufcient motivation to proceed with equation (6) as an adequate, and much
more practical, representation of equation (2).

3. PERTURBATION ANALYSIS

The method of multiple scales [3739] is now so well established that little background
explanation is needed other than to conrm that it uses the concept of an innite series
solution with successively higher order terms in the series offering non-linear corrections
to the principal part, normally the lowest order linear solution. In this case the series is
truncated after the rst order term in e; a decision forced by the absence of terms to higher
order e in either of the differential equations (5) and (6). Therefore the solutions for x and

923

NON-LINEAR METAL CUTTING MODEL

1, 2

0.5

fx

fy

vf

vr
0
-2

-1

-2

-1

-1

-0.5

(a)

2.5

(b)

1.5

fx

fy
2

1, 2

1
1.5

1
0.5
0.5

0.4

0.8

1.2

1.6

0
2

0.2

(c)

0.4

0.6

0.8

(d)

Figure 2. (a) Plot of force fx as a function of the relative velocity between the tool and the work-piece in the x
direction, vr ; for the full force term as stated in the right-hand side of equation (1)}curve 1, and the
approximation as calculated from the right-hand side of equation (5)}curve 2 (approximation and full term are
indistinguishable). Additional data is, m0 01; C1 03; C2 07; C3 15; C4 12; h0 05; R0 22; a 4:
(b) Plot of force fy as a function of the relative velocity between the tool and the work-piece in the y direction, vf ;
for the full force term as stated in the right-hand side of equation (2)curve 1, and the approximation as
calculated from the right-hand side of equation (6)curve 2. Additional data is, m0 01; C1 03; C2 07;
C3 15; C4 12; h0 05; R0 22; a 4: (c) Plot of force fx as a function of the actual cutting depth, h; for
the full force term as stated in the right-hand side of equation (1)}line 1, and the approximation as calculated
from the right-hand side of equation (5)}line 2 (approximation and full term are indistinguishable). Additional
data is, m0 01; C1 03; C2 07; C3 15; C4 12; h0 05; R0 22; a 4: (d) Plot of force fy as a
function of the actual cutting depth, h; for the full force term as stated in the right-hand side of equation (2)curve
1, and the approximation as calculated from the right-hand side of equation (6)curve 2. Additional data is,
m0 01; C1 03; C2 07; C3 15; C4 12; h0 05; R0 22; a 4:

y are given by
x x0 ex1 . . . ;

y y0 ey1 . . . :

7; 8

Following the method the rst and second time derivatives are also perturbed as follows:
d
@
@

e
. . . D0 eD1 . . . ;
dt @T0
@T1

d2
D20 2eD0 D1 . . . ;
dt2

9; 10

924

J. WARMINSKI ET AL.

The perturbed derivatives explicitly introduce new independent variables (timescales), T0

and T1 ; such that x xT0 ; T1 and y yT0 ; T1 and the subscripts 0 and 1 are used in
the perturbation equations to show appropriate dependency on these two timescales.
Separation of terms to Oe0 and Oe1 leads to two perturbation equations for x and also
for y; thus,
0e0 :
0e1 :

D20 x0 x0 Q2  Q1 y0 ;

11

D20 x1 x1  2D0 D1 x0  2nx g1 D0 x0  g2 y0

g3 D0 x0 2  g4 D0 x0 2 y0 g5 D0 x0 y0  Q1 y1 ;
0e0 :

0e1 :

D20 y0 a Q3 y0 l11 ;

12
13

p
D20 y1 a Q3 y1  2D0 D1 y0 2ny a l1 D0 y0
 P% 1 y0 l2 D0 y0 2  l3 D0 x0 2 y0  l4 D0 x0 y0  l5 y3
0

 l6 D0 y0 y0  l7 D0 y0 y0
l8 D0 x0 2 l9 D0 x0 l10 y20 P2 ;

14

where
o2x 1;

o2y a Q3 ;

n1 nx ;

p
n 2 n y a;

It can be readily seen that the zeroth order perturbation equations (11) and (13) are
inhomogeneous, but they are linear equations and can therefore be solved using
conventional ordinary linear differential equation theory. In fact, it is simpler to start with
equation (13) so that this solution for y0 can then be used to construct the solution for x0 :
The solutions for x0 and y0 are as follows, given here in complex exponential form,
x0 A1 eiox T0 A1 eiox T0 K1 A2 eioy T0 A2 eioy T0 K2 K1 K3 ;

15

y0 A2 eioy T0 A2 eioy T0 K3 ;

16

where
K1

o2x

Q1
;
 o2y

K2

Q2 K1 l11

;
o2x
o2x

K3

l11
:
o2y

Complex amplitudes A1 and A2 can be stated in terms of amplitudes a1 and a2 and phases
f1 and f2 by means of equations (17) and (18)
A1 12 a1 eif1 ;

A2 12 a2 eif2 ;

17; 18

From this the zeroth order perturbation solutions can be recast in trigonometrical form, if
required at this stage,
x0 a1 cosox T0 f1 K1 a2 cosoy T0 f2 K2 K1 K3 ;

19

y0 a2 cosoy T0 f2 K3 :

20

For the purposes of generating the rst order perturbation corrections (i.e., x1 and x2 ) it is
more convenient to continue using the zeroth order perturbation solutions in the form of
equations (15) and (16). These are appropriately substituted into equations (12) and (14)
and then resonance conditions are identied from the exponents of the resulting terms in

NON-LINEAR METAL CUTTING MODEL

925

the usual manner. The most interesting result is the non-resonant case because this is
found between all the various resonant cases and must therefore occur for the widest range
of machining conditions. Five internal resonance conditions can also be readily identied
as, oy  3ox , oy  2ox ; oy  ox ; ox  2oy ; and ox  3oy : There are no external
resonances because this is a self-excitation problem that does not involve explicit external
excitations. Because the non-resonant case is obviously of very particular practical interest
it is examined exclusively in the remainder of this paper. In order to get to this point a
more detailed consideration of the rst order perturbation equations is necessary.
Returning to equations (12) and (14), the two zeroth order solutions can be substituted in
(from equations (15) and (16)) and differentiations with respect to T0 carried out at the
same time, to lead to new forms for the rst order perturbation equations. The two rst
order perturbation equations contain very large numbers of terms and are therefore not
shown here due to space restrictions. In fact specialized symbolic computer software [33,
34] has been employed throughout in order to do the complete multiple scales analysis.
This is implemented within the Mathematica computer algebra system in the form of a new
package called MultipleScales and has ensured the accuracy of the solution. All the
principal results for the non-resonant case of interest are quoted from the formal output
generated by this semi-automated approach.
Therefore, by continuing to apply the method of multiple scales all possible resonance
conditions can be identied by inspection of the expanded right-hand sides of the rst
order perturbation equations. This leads to the immediate identication of secular terms
(i.e., those terms that will invalidate the uniformity of the power series solutions to x and
y). As stated above, the fully expanded forms of the rst order perturbation equations
are not quoted explicitly, however the secular terms are shown for the interesting
non-resonant condition, and are dened in the paper as solvability conditions, according to
the accepted terminology of the method. From this it is then shown how the
particular solutions to x1 and y1 can be constructed for the stable non-resonant
case, on the understanding that this is the trivial solution from the point of view of the
vibration analysis but the most desirable condition from the point of view of the
machining process. The complete solutions for x and y under the stable non-resonant
conditions are subsequently obtained by using the original series solutions given in
equations (7) and (8).
3.1. ANALYSIS OF THE NON-RESONANT CASE
The solvability conditions for this case are obtained from the right-hand sides of the rst
order perturbation equations and are as follows:
iA1 ox g1 o2x  K3 g5 o2x 2n1 o2x  g1 o2y K3 g5 o2y  2n1 o2y K3 Q1 l4  Q1 l9
2iox A01 0;
ox  oy ox  oy
21
A% 2 K12 l3 o2y l7 o2y 3l5 A22 3l5 K32  2l10 K3 iK2 l4 oy K3
il6 oy K3 2A% 1 A1 l3 o2x P1  il1 oy  iK1 l9 oy 2in2 oy A2 2ioy A02 0; 22
noting that they are still in terms of complex amplitudes A1 and A2 : Substitution of
equations (17) and (18) for these complex amplitudes in equations (21) and (22), and then
separation of terms into real and imaginary parts, yields four modulation equations, as
stated in Appendix A, (equations (A.1)(A.4) inclusive). These modulation equations are
readily solved and lead to the following solutions for a1 ; a2 ; f1 ; and f2 ; noting that c1 ;

926

J. WARMINSKI ET AL.

c2 ; and c3 are constants of integration,

K 3 Q 1 l4
Q 1 l9

a1 c3 exp 
2ox  oy ox  oy 2ox  oy ox oy
g1 o2x
K3 g5 o2x
n1 o2x


2ox  oy ox oy 2ox  oy ox oy ox  oy ox oy
!
g1 o2y
K3 g5 o2y
n1 o2y

t;
2ox  oy ox oy 2ox  oy ox oy ox  oy ox oy


23


l1 1
K3 l6 K1 l9
 K 1 K 3 l4 

 n2 t;
a2 c2 exp
2 2
2
2

24

f1 f1 0;

25

2
P1 3K32 l5  2K3 l10 t el1 K1 K3 l4 K3 l6 K1 l9 2n2 t c2 3l5 K12 l3 o2y l7 o2y

2oy
8l1  K1 K3 l4  K3 l6 K1 l9  2n2 oy
!
2
K3 Q1 l4  Q1 l9 g1 ox  K3 g5 o2x 2n1 o2x  g1 o2y K3 g5 o2y  2n1 o2y t
exp
ox oy ox oy

f2 c1

c3 2 l3 o2x ox oy ox oy =
4oy K3 Q1 l4  Q1 l9 g1 o2x  K3 g5 o2x 2n1 o2x  g1 o2y K3 g5 o2y  2n1 o2y :

26

The constants of integration can be determined from the initial displacement and velocity
conditions for x and y; however because of the particular structure of equations (23)(26)
it is much easier to solve the inverse problem in which one proceeds from assumed
constants to calculate corresponding initial conditions. An iterative calculation can then
be used to determine the constants for a desired set of initial conditions. It is important to
appreciate that as time approaches innity the amplitudes a1 and a2 will tend to zero if the
overall value of the constant part of the exponents of equations (23) and (24) is negative.
In other words, if


K3 Q1 l4  Q1 l9 g1 o2x  K3 g5 o2x 2n1 o2x  g1 o2y K3 g5 o2y  2n1 o2y

2ox  oy ox oy

50:

27

and
1
2l1

 K1 K3 l4  K3 l6 K1 l9  2n2 50:

28

Inequalities (27) and (28) do not contain the constants of integration and are therefore
independent of the initial conditions, as, therefore, are the conditions for stability. If
inequalities (27) and (28) are not satised then the amplitudes will grow exponentially with
time, and are denitionally unstable. Assuming that inequalities (27) and (28) are satised,
and this implies conditions for stability, then the particular solutions to the rst order
perturbations can be obtained, notwithstanding the somewhat tedious algebra required.
This is also handled by the symbolic solution software [33, 34], to give
x1

P2 Q1 K33 Q1 l5  K32 Q1 l10 K3 P1 Q1  g2 o2y

;
o2x o2y

29

K3 P1 P2  K33 l5 K32 l10

:
o2y

30

y1

NON-LINEAR METAL CUTTING MODEL

927

These solutions for x1 and y1 ; together with those for x0 and y0 ; can be substituted into
equations (7) and (8) to give the complete solutions for x and y for the non-resonant case,
on the understanding that inequalities (27) and (28) hold true. The solutions for x0 and y0
will obviously equal (K2 K1 K3 ), and K3 ; respectively, given that a1 1 a2 1 0 in
this case. From this it can be seen that the stable, non-resonant, solutions for x and y are
constant valued dc offsets. These are the trivial solutions for the vibration analysis, as
discussed above, but highly desirable as far as the machining process is concerned. In
order to examine the effect of these conditions in practice inequalities (27) and (28) can be
restated in terms of the original physical parameters, thus,
p
p
C1 h0 q20 1 v0 a  a1 q0 aC2 h0 q0 R0 1 v0 a q0 axy
50;
31
e1 q0 aq0 a
C1 h0 q20 1 v0 a  av0 a q0  q20 a  2q0 a  a2 xx
p
p
50;
e1 q0 a1 q0 aq0 a

32

noting that these are for full contact conditions dened by sgn( )=1 and H( )=1, in the
limit of relatively small oscillations. If inequalities (31) and (32) are satised then they can
be rearranged to show that the work-piece velocity, v0 ; must satisfy both the following
conditions for stability:
p
p
C1 h0 q20 a  a1 q0 aC2 h0 q0 R0 a q0 axy
;
33
v01 > 
C1 h0 q20 a  C2 h0 q0 R0 a1 q0 a
C1 h0 q20 1 aa  q0 a1  q0 axx
:
v02 > 
C1 h0 q0 1  aa

34

Figures 3(a) and 3(b) show plots of v0 as functions of cutting force magnitude, q0 ; and
initial depth of cut, h0 ; with the upper solid line delineating the theoretically stable region
(white) from the theoretically unstable region (grey). The dotted line is the commensurate
numerical solution for the upper solid line. Numerical simulations were carried out by
means of the fourth order Runge Kutta method, and vibrations of the system were
analyzed for small initial conditions. The dotted line in the gures divides the parameter
plane into two regions with a stable regime predicted for the space above the line and
unstable behaviour below. It can be seen in Figure 3(a) that there is close qualitative
similarity between them but a fairly constant quantitative discrepancy of approximately
20%. This is because the numerical solutions for v0 (versus q0 and h0 ) have been obtained
from the original full equations (1) and (2), whereas the conditions in inequalities (33) and
(34) emanate from the multiple scales analysis. As is already apparent the multiple scales
analysis contains inherent approximations due to series truncation and the assumptions
that accompany the use of solvability conditions in general, in addition to the special
modelling simplications that took place at the time of term ordering so as to ensure a
reasonably tractable algebraic problem from thereon. This discrepancy is not consistent,
or as signicant, in Figure 3(b).
It is also instructive to examine the zeroth order perturbation solutions, x0 and y0 in the
time domain, and two cases for each are presented in Figures 4 and 5. In Figure 4 solutions
are given for x0 and y0 for a selected stable point as dened in Figure 3(b), where q0 1;
h0 05; and v0 08 (i.e. a point within the white region in Figure 3(b)). It can be seen
that both responses decay with time to what will become levelled-out dc offsets a little later
on. The rst order perturbation corrections of equations (29) and (30) would add small
corrections to these two responses. The converse situation is shown in Figure 5 where the
operating reference point in Figure 3(b) is shifted into the unstable grey region, such that

928

J. WARMINSKI ET AL.
1

NS
stable

NS
0.8

stable

0.8

AR
0.6

v0

v0

0.6
AR

unstable

0.4

0.4
unstable

0.2

0.2

0
0

(a)

0.4

0.8

1.2

q0

1.6

0.4

(b)

0.8

1.2

1.6

h0

Figure 3. (a) Plot of feed velocity vo as a function of the magnitude of the cutting force q0 ; for the multiple
scales analysis (analytical result AR}solid curve) and for numerical analysis based on the original equations of
motion due to Wiercigroch [23] (numerical solution NS}dotted curve). Stable area is white, unstable area is grey.
Inner analytical solution is not necessary for the stability criterion. Additional data is, m0 01; C1 03; C2
07; C3 15; C4 12; h0 05; R0 2:2; a 4; xx 01; xy 01: (b) Plot of feed velocity vo as a function of
the initial depth of cut h0 ; for the multiple scales analysis (analytical result AR}solid curve) and for numerical
analysis based on the original equations of motion due to Wiercigroch [23] (numerical solution NS}dotted
curve). Stable area is white, unstable area is grey. Inner analytical solution is not necessary for the stability
criterion. Additional data is, m0 01; C1 03; C2 07; C3 15; C4 12; R0 22; a 4; xx 01; xy 01:

q0 1; h0 05; and v0 04: The time-domain plots for x0 and y0 are given in Figure 5

and show typical exponential growth with time, indicating instability. It is worth noting
that the time response for x0 is also subjected to a low frequency modulation, this being an
additional, unexplored, interaction between the work-piece feed speed v0 and the local selfexcitation due to the orthogonal interactions in x and y:

4. SUMMARY AND CONCLUSIONS

This work has shown that it is possible to generate approximate analytical solutions to
the non-resonant, self-excited, orthogonal metal cutting problem, as represented in a twodegree-of-freedom system in which dry friction chatter occurs. The multiple scales method
as applied, has enabled an approximate analysis to be undertaken of the dynamic
interactions occurring in the metal cutting process. This facilitates the provision of
guidance on the selection of technological parameters for the avoidance of primary
chatter. This paper looks, for the rst time, at the predicted stability regions on the basis of
a practically useful technological parameter space, involving the depth of cut and the
cutting speed. The non-resonant case is of particular interest because it is necessarily a
much more generally widespread, and therefore potentially dangerous, phenomenon than
the specic resonances that are also predicted by the analysis. The solutions for the nonresonant case have been generated from structurally simplied equations of motion based
on the seminal equations rst postulated by Wiercigroch [23], although much of the
original discontinuous behaviour of Wiercigrochs model is still retained within the
simplied equations. The simplications extend to a reduction in the complexity of the

929

NON-LINEAR METAL CUTTING MODEL

x0
1.5
1
0.5

10

20

30

40

50

10

20

30

40

50

60

0.5
-1
1.5
(a)

y0
1
0.75
0.5
0.25
60

0.25
-0.5
0.75
-1
(b)
Figure 4. Time domain plots for x0 (a) and y0 (b) for stable cutting, for which q0 1; h0 05; and v0 08:
Additional data is, m0 01; C1 03; C2 07; C3 15; C4 12; R0 22; a 4; xx 01; xy 01: The
integration constants used to calculate the time domain responses are C1 C2 C3 1; with f0 0;
which equate to initial conditions x0 143; x 0 091; y0 00065; y 0 348:

terms expressing the orthogonal cutting forces, fx and fy : Although there are no discernible
differences between the full and simplied forces in the x direction, and little difference
between their counterparts in the y direction, there is around 20% difference between the
feed-rate velocities predicted by the multiple scales analysis and those calculated
numerically from the original equations of motion due to Wiercigroch [23]. This
difference is largely attributable to an accumulation of errors starting from the structural
simplications to the equations of motion and then through using the approximate
perturbation method of multiple scales. Regarding the specic resonance conditions that
are also predicted by the multiple scales analysis, these involve relationships in the form of
positive integers between the natural frequencies ox and oy which lead to additional forms
of localised resonant behaviour.
The multiple scales perturbation analysis was carried out by using a specialized
symbolically based computational solver, Khanin et al. [33] and Khanin and Cartmell [34],
and has led to an analytical criterion, in the form of formulas (31) and (32), for the
denition of stable, or unstable, dry friction chatter vibrations in orthogonal metal
cutting. Compared with numerical methods, including neural network procedures, this is a

930

J. WARMINSKI ET AL.

x0
6
4
2

t
2.5

7.5

10

12.5

15

17.5

20

2.5

7.5

10

12.5

15

17.5

20

-2
-4
-6
(a)

Y0
20
15
10
5

t
-5
-10
-15
-20
(b)
Figure 5. Time domain plots for x0 (a) and y0 (b) for unstable cutting, for which q0 1; h0 05; and
v0 04: Additional data is, m0 01; C1 03; C2 07; C3 15; C4 12; R0 22; a 4; xx 01; xy 01:
The integration constants used to calculate the time domain responses are C1 C2 C3 1; with f0 0;
which equate to initial conditions x0 143; x 0 091; y0 00065; y 0 348:

fast and elegant approach to such a criterion, although it is clear that a good dynamical
model and physically realistic data are important pre-requisites.
It is also shown (in Figure 3(b)) that dry friction chatter stabilises for deep cutting at
high feed velocities but destabilizes at low feed velocities.
There is also an, as yet, unexplored low-frequency modulation of the theoretically
predicted time domain response in the x direction for the unstable case, but it is thought
that this is due to dynamic coupling between the work-piece feed and the local dynamics of
the cutting tool, as modelled here.
Further work is planned in the form of experiments to be carried out at the Technical
University of Lublin to determine stability/instability transitions against which the
analytical predictions stated within this paper can be checked for a range of cutting
parameter data. This will determine to what extent the analysis could be revised for higher
levels of precision. The low frequency modulation that is predicted for the response of the
tool in the x direction in the unstable case is also to be investigated and reported in the
literature in due course. The resonant phenomena predicted by the analysis will also be

NON-LINEAR METAL CUTTING MODEL

931

analyzed and reported as appropriate. A new programme of research on the use of the
results in this paper for the real-time control of the cutting process is under consideration,
and this relates to the recent work of Pratt and Nayfeh [40] on chatter control and stability
analysis of a cantilever boring bar under regenerative cutting conditions. It is also intended
that the work of this paper could be used to provide a map of stable cutting parameters for
workshop use in the manual setting-up of machine-tools.
The results from this research should be adaptable to other machining processes such as
turning or milling, and new work is planned to explore these other applications.

ACKNOWLEDGMENTS

The authors would like to acknowledge the nancial support provided by the Polish
State Committee for Scientic Research, grant number 126/E-361/SPUB/COST/T-7/DZ
42/99, the Polish Academy of Sciences, the British Council, and the Royal Society, for
funding the collaborations which led to the completion of this work. They would also like
to note the support provided by the UKs Engineering and Physical Sciences Research
Council on grant GR/L30749 for the independent research which led to the development
of the semi-automated symbolic solver package used for the multiple scales analysis in this
research programme. The Royal Society Dorothy Hodgkin Research Fellowship awarded
to R. Khanin at the University of Glasgow and then at the University of Cambridge, is
also gratefully acknowledged. The authors would also like to thank Eng. B. Paw"owska
for her help in compiling the manuscript.

REFERENCES
1. S. A. Tobias 1965 Machine Tool Vibration. New York: Wiley.
2. H. Opitz and F. Bernardi 1970 Annals of the CIRP 18, 335343. Investigation and calculation
of the chatter behaviour of lathes and milling machines.
3. J. Tlusty 1986 American Society of Mechanical Engineers, Journal of Engineering for Industry
108, 5967. Dynamics of high speed milling.
4. I. Grabec 1986 Physics Letters A 117, 384386. Chaos generated by the cutting process.
5. I. Grabec 1988 International Journal of Machine Tool Manufacture 28, 1931. Chaotic dynamics
of the cutting process.
6. E. Marui, S. Kato and M. Hashimoto 1988 American Society of Mechanical Engineers, Journal
of Engineering for Industry 110, 236241. The mechanism of chatter vibration in a spindleworkpiece system. Part 1}properties of self-excited chatter vibration in a spindle-workpiece
system 110, 242247. The mechanism of chatter vibration in a spindle-workpiece system. Part
2}characteristics of dynamic cutting force and vibration energy 110, 248253. The mechanism
of chatter vibration in a spindle-workpiece system. Part 3}analytical considerations.
7. J. S. Lin and C. I. Weng 1990 International Journal of Machine Tool Manufacture 19, 5364. A
non-linear dynamic model of cutting.
8. J. S. Lin and C. I. Weng 1991 International Journal of Mechanical Sciences 33, 645657. Nonlinear dynamics of the cutting process.
9. J. Gradisek, E. Govekar and I. Grabec 1996 International Journal of Machine Tool
Manufacture 36, 11611172. A chaotic cutting process and determining optimal cutting
parameter using neural networks.
10. J. Gradisek, E. Govekar and I. Grabec 1998 Mechanical Systems and Signal Processing 12,
839854. Time series analysis in metal cutting: chatter versus chatter-free cutting.
11. J. Gradisek, E. Govekar and I. Grabec 2001 Journal of Sound and Vibration 242, 829838.
Chatter onset in non-regenerative cutting: a numerical study.
12. G. Stepan and T. Kalmar-Nagy 1997 Proceedings of DETC 97, 16th ASME Biennial
Conference on Mechanical Vibrations and Vibrations and Noise, September 1417, Sacramento,
CA, U.S.A., 111. Non-linear regenerative machine tool vibrations.

932

J. WARMINSKI ET AL.

13. G. Litak, J. Warmin ski and J. Lipski 1997 Proceedings of the 4th Conference on Dynamical
Systems}Theory and Applications, o! d"z, Poland, 89 December, (J. Awrejcewicz, J. Grabski
and J. Mrozowski, editors), 193198. Self-excited vibrations in the cutting process.
14. W. Przystupa and G. Litak 1998 Proceedings of the XXXVII Symposium on Modelling in
Mechanics, Wis!a, Poland, 305310. Cutting process in the presence of noise.
15. M. A. Johnson and F. C. Moon 2001 International Journal of Bifurcation and Chaos 11,
449467. Non-linear techniques to characterise pre-chatter and chatter vibrations in the
machining of metals.
16. G. Stepan 2001 Philosophical Transactions of the Royal Society of London A 359, 739757.
Modelling nonlinear regenerative effects in metal cutting.
17. M. Wiercigroch and E. Budak 2001 Philosophical Transactions of the Royal Society of
London A 359, 663693. Sources of non linearities, chatter generation, and suppression in metal
cutting.
18. F. C. Moon, and T. Kalmar-Nagy 2001 Philosophical Transactions of the Royal Society of
London A 359, 695711. Non-linear models for complex dynamics in cutting materials.
19. E. Rabinowitz 1958 Proceedings of the Physical Society 71, 668673. The intrinsic variables
affecting the stick-slip process.
20. R. Bell and M. Burdekin 1970 Proceedings of the Institution of Mechanical Engineers 184,
543557. A study of stick-slip motion of machine tool drives.
21. M. Wiercigroch 1989 Ph.D. Thesis, Silesian Technical University, Gliwice, Poland. Modelling
of dynamic interaction between machine tool and cutting process (in Polish).
22. M. Wiercigroch and J. Kosmol 1993 GEP, XLV, 3941. Active control of nonlinear
vibrations of the MTCP system.
23. M. Wiercigroch 1997 Transactions of the American Society of Mechanical Engineers, Journal of
Vibration and Acoustics 119, 468475. Chaotic vibrations of a simple model of the machine
tool}cutting process system.
24. M. Wiercigroch and A. H.-D. Cheng 1997 Chaos, Solitons, and Fractals 8, 715726. Chaotic
and stochastic dynamics of orthogonal metal cutting.
25. J. Warmi ski, G. Litak, J. Lipski, M. Wiercigroch and M. P. Cartmell 2000 in IUTAM/
IFToMM Symposium on Synthesis of Nonlinear Dynamical Systems (E. Lavendis and
M. Zakrzhevsky, editors.), Dordrecht, Netherlands: Kluwer Academic Publishers. 275284.
Chaotic vibrations in the regenerative cutting process.
26. T. Kalmar-Nagy and J. R. Pratt 1999 Proceedings of DETC 99, 17th Biennial Conference on
Mechanical Vibration and Noise, Las Vegas, U.S.A., DETC99/VIB-8060. Experimental and
analytical investigation of the subcritical instability in metal cutting.
27. M. Wiercigroch and A. Krivtsov 2001 Philosophical Transactions of the Royal Society of
London A 359, 713738. Frictional chatter in orthogonal metal cutting.
28. E. P. Nosyreva and A. Molinari 1998 International Journal of Mechanical Sciences 40,
735748. Analysis of non-linear vibrations in metal cutting.
29. M. E. Merchant 1944 Transactions of the American Society of Mechanical Engineers, Journal of
Applied Mechanics A 11, 168175. Basic mechanics in the metal cutting process.
30. C.-C. Hwang, R.-F. Fung and J. S. Lin 1997 Journal of Sound and Vibration 203, 363372.
Strong non-linear dynamics of cutting process.
31. N. H. Hanna and S. A. Tobias 1974 Transactions of the American Society of Mechanical
Engineers, Journal of Engineering for Industry 96, 247255. A theory of nonlinear regenerative
chatter.
32. A. H. Nayfeh and C.-M. Chin and J. Pratt 1997 Journal of Manufacturing Science and
Engineering 195, 485493. Perturbation methods in nonlinear dynamics}applications to
machining dynamics.
33. R. Khanin, M. P. Cartmell and A. D. Gilbert 2000 Computers and Structures 76, 565575. A
computerised implementation of the multiple scales perturbation method using Mathematica.
34. R. Khanin and M. P. Cartmell 2001 Journal of Symbolic Computing 31, 461473.
Parallelisation of perturbation analysis: application to large-scale engineering problems.
35. F. Jiulian, Y. Zhejun and Y. Yingxue 1989 International Journal of Machine Tool
Manufacture 29, 601609. A unied system model of cutting chatter and its transformation
function.
36. W. F. Hastings, P. Mathew and P. L. B. Oxley 1980 Proceedings of the Royal Society of
London A 371, 343354. A machining theory for predicting chip geometry, cutting forces, etc.
from material properties and cutting conditions.

NON-LINEAR METAL CUTTING MODEL

933

37. A. H. Nayfeh 1974 Perturbation Methods. New York: Wiley.

38. A. H. Nayfeh and D. T. Mook 1979 Nonlinear Oscillations. New York: Wiley.
39. M. P. Cartmell 1990 An Introduction to Linear, Parametric and Nonlinear Vibrations. London:
Chapman & Hall.
40. J. R. Pratt and A. H. Nayfeh 2001 Philosophical Transactions of the Royal Society of London A
359, 759792. Chatter control and stability analysis of a cantilever boring bar under regenerative
cutting conditions.

APPENDIX A

The modulation equation for the non-resonant case are as follows:

a1 ox f01 0;
a32

3l5 K12 l3 o2y l7 o2y

8
8
8

A:1

!
14 a2 2P1 6K32 l5  4K3 l10 a21 l3 o2x  4oy f02 0;

A:2

a1 ox K3 Q1 l4  Q1 l9 g1 o2x  K3 g5 o2x 2n1 o2x  g1 o2y

K3 g5 o2y  2n1 o2y =2ox  oy ox oy ox a01 0;

A:3

12 a2 l1  K1 K3 l4  K3 l6 K1 l9  2n2 oy oy a02 0:

A:4